About the mask in the training process of Diffusion

[xiutianli]

Thank you for good implementation of Palette! Forgive my unprofessional question, after reading your code, but I want to know that whether the mask keeps the same during the training process of diffusion, since it's loaded in dataloader?